Skip to main content

Sueperman

For everyone who's ever bought something, forgotten the receipt, and never realized they might be owed a remedy… one unlikely hero has their

About the project

SUEPERMAN

In a world where every purchase tells a story… some stories come with a warning.

Products recalled. Settlements forgotten. Refunds buried beneath mountains of paperwork.

And millions of receipts gathering dust, their secrets waiting to be discovered.

But when the system gets complicated… an unlikely hero rises.

Introducing… SUEPERMAN!

Armed with nothing but your purchase history and an extraordinary sense of justice, he searches official recalls, investigates hidden opportunities, verifies the evidence, and prepares the paperwork—so you don’t have to fight the bureaucracy alone.

No super strength. No laser vision. No guaranteed refunds.

Just one extraordinary superpower: knowing what your receipts might be worth.

This isn’t the hero Metropolis asked for…

It’s the hero your purchase history needs.

SUEPERMAN

Your receipts have unfinished business.

Built with

pythonjavascriptopenaifastapi

AI tools

Coded withGPT-6.1 Sol

Prize Categories Entered

  • Gemini Prize Category Q1 2025 - Q1 2026

    Best Use of Gemini API

    Submitted

    Sueperman uses Google Gemini to turn receipt images and PDFs into structured purchase records, extracting merchant, purchase date, product names and visible identifiers. We validate its output against typed schemas and reject malformed results. Extracted identifiers remain unverified suggestions until the user reviews and confirms them. This connects document understanding to an actionable consumer workflow: purchases become inputs for source-backed recall and settlement checks. We verified live PNG and PDF extraction, capped generation output, and reused duplicate imports to avoid unnecessary API spending.

  • DigitalOcean

    Best Use of DigitalOcean

    Submitted

    Sueperman runs on DigitalOcean App Platform, which builds and deploys our Dockerized Next.js frontend and FastAPI backend from GitHub. Its HTTPS ingress serves the application and routes /api requests to the backend. We use API-service runtime secrets to protect the OpenAI credential powering receipt-photo reading, and configured health checks to monitor backend availability. We verified the deployed application through browser workflows covering receipt extraction, source-backed recall and settlement checks, and local draft approval/export.

  • Sentry

    Sentry Raffle

    Submitted

    We integrated Sentry across Sueperman’s Next.js frontend and FastAPI backend to capture errors, performance traces, operational logs and browser replays. Because receipts can contain sensitive information, we disabled request-body and AI-input/output collection, scrubbed credentials and exception details, and masked replay text and inputs. We retained error types, source locations and correlation IDs so failures remained diagnosable. We verified the integration by triggering real browser and backend failures, confirming delivery in Sentry, inspecting masked replay data, and uploading source maps to verify readable application stack frames. Privacy tests caught leaks and missing diagnostic metadata, which we repaired before removing the temporary error triggers.

  • Bright Data

    Best Use of Bright Data

    Submitted

    We used Bright Data’s official MCP server to extend government recall records with manufacturer remedy instructions. Sueperman runs search_engine, then scrape_as_markdown against a supported CPSC-linked manufacturer page. It stores relevant model/refund excerpts, source URLs, timestamps and content hashes, and displays them alongside the official evidence. We verified both tools live against NEWDERY’s recall page and confirmed the manufacturer citation appeared in the browser workflow. Scraped content stays untrusted data; it cannot override deterministic eligibility checks or trigger claim submissions.

  • HyperFrames

    Best Demo Video Using HyperFrames

    Submitted

    We used HyperFrames to author and render Sueperman’s 90-second superhero-style product commercial. The editable project combines illustrated motion-comic scenes, timed dialogue, animated typography, transitions, music and sound effects with actual application screenshots demonstrating recall matching, near-match rejection and local draft approval/export. HyperFrames rendered all 2,700 frames into a 1920×1080, 30 FPS MP4 with audio. Delivery checks verified duration, resolution, frame rate and complete decoding. Optional captions are included. The video preserves replay/simulation labels and avoids fabricated testimonials, payouts or claim-submission results.

  • Event prize

    Best Overall Hack

    Submitted

    Sueperman turns an everyday artifact—a receipt—into useful, source-backed next steps. Users can photograph a receipt, review extracted purchases, discover recall and settlement opportunities, inspect official evidence, and prepare a locally approved remedy draft. The project combines a complete consumer workflow with deliberate engineering: live integrations, exact-model verification, visible near-match rejection, explicit uncertainty, privacy-conscious monitoring and spending controls. It also has a deployed public application, a reproducible demo and tested browser workflows. Its strongest distinction is that the evidence stays visible throughout the experience. Users can understand what matched, what remains unknown and what they must verify before acting. That makes the hack useful beyond the initial AI demonstration.

  • Event prize

    Best Use of System One Models

    Submitted

    Sueperman presents a concrete decision problem: given a purchase, source evidence and missing information, what should the investigation do next? Our OpenAI-powered router chooses from four bounded actions: inspect more evidence, request user information, prepare a packet or reject a candidate. Each choice has an observable consequence, and the UI shows the selected action and decision evidence. Deterministic eligibility gates remain authoritative, so a model cannot turn a near-match or missing identifier into a verified claim. This makes decision quality testable through real outcomes—especially false-positive rejection and safe handling of uncertainty.